Integrating chronological aging and asynchronous aging for enhanced biological age prediction using artificial...

The accurate estimation of biological age (BA) is limited by the reliance on chronological age (CA), which does not account for asynchronous aging-heterogeneity in individual aging trajectories. To address this, we propose a unified artificial intelligence framework that explicitly integrates CA with a quantifiable measure of asynchronous aging for BA prediction. We quantify asynchronous aging index (AAI) using a pre training framework. The AAI is defined as the deviation of an individual's predicted age from the average predicted age of a healthy reference population-a concept that has been validated in large-scale proteomic aging studies. Moreover, we propose and evaluate three distinct strategies to combine CA and AAI in re-training framework: AAI-score, a direct composite measure; Loss(AAI,MSE), a hybrid loss function; AAI-driven data cleaning procedure. Applied to an arterial stiffness data from over 36,000 individuals, our integrated approaches uniformly enhance the predictive accuracy of traditional basic framework without considering asynchronous aging. The most pronounced improvement is achieved by the AAI-score, which reduces the MAE in males from 7.01-7.28 years to 2.99-3.69 years, and in females from 5.83-6.09 years to 3.51-4.05 years. Concurrently, the AUC for health risk classification rises from 0.34-0.68 to 0.66-0.89 in males and from 0.28-0.50 to 0.60-0.90 in females. Our work establishes that asynchronous aging is a fundamental component of BA. By formally integrating AAI with CA within BA prediction framework, we provide a more accurate, interpretable, and clinically promising path for BA estimation.